EECS 750. Hypothesis Testing with Communication Constraints
|
|
- Joshua Bryan
- 5 years ago
- Views:
Transcription
1 EECS 750 Hypothesis Testing with Communication Constraints Name: Dinesh Krithivasan Abstract In this report, we study a modification of the classical statistical problem of bivariate hypothesis testing. The statistician has to make his decisions based on remotely collected data available to him through a noiseless channel having a finite rate. The problem is to determine the minimum β n of the second kind of error probability under the condition that the first kind of error probability α n ɛ for a given 0 < ɛ < 1. We first establish the basic results using techniques of information theory. We then derive a single-letter lower bound for the general bivariate testing case in the presence of one-sided data compression. Special cases of this bound for the case of test against independence and extensions to two-sided data compression are discussed.
2 1 Introduction One of the most fundamental and standard problems in statistics is to decide which one of the two available explanations best explains the observed data x n = (x 1, x 2,..., x n ). For example, based on n measurements of sensor data, one might want to determine whether an earthquake occurred or not. In the simplest hypothesis testing problem, we assume that the data is generated i.i.d from either the distribution P = (P (x)) x X (Hypothesis H 0 ) or from the distribution Q = (Q(x)) x X (Hypothesis H 1 ). The statistician has to decide between H 0 and H 1 based on a sample of size n. The most general decision rule is to declare H 0 if x n A X n and declare H 1 otherwise. There are two kinds of errors involved in this problem. A type-1 error is said to occur if statistician declares H 1 when in fact the data was generated according to H 0. Type-2 error is similarly defined. From the decision rule given above, it can be seen that P ( Error of type 1 ) α n = P n (A c ) P ( Error of type 2 ) β n = Q n (A) It is clear from the above definition of α n and β n that there is a tradeoff between them. 1.1 Neyman-Pearson Lemma The Neyman-Pearson lemma derives the optimal test between two hypotheses (i.e.) it gives the optimal way to choose the set A. The lemma can be stated as follows. Lemma 1.1. For T 0, define the region { A n (T ) = x n : P n (x n } ) Q n (x n ) > T. Then no other decision region B n X n can be found that has lower error probabilities for both type-1 and type-2 error events than that of the set A n. In other words, the optimal test is of the form P (x n ) Q(x n ) T 1
3 where the choice of T determines the tradeoff between α n and β n 1.2 Stein s Lemma Very often, H 1 denotes a highly critical event (such as the occurrence of an earthquake) while H 0 denotes its negation ((i.e) non-occurrence of the earthquake). Consequently type-2 error events are more critical than type-1 error events. Thus, we want β n to decay to 0 as fast as possible while we only want that α n go to 0 as n goes to. We need to find A n X n with P n (A n ) 1 ɛ and Q n (A n ) = β n (ɛ) where 0 < ɛ < 1 is given and β n (ɛ) min A { Qn (A n ) A X n, P n (A) > 1 ɛ} The rate of convergence of β n (ɛ) as n goes to is given by Stein s Lemma which is stated below. Lemma 1.2. For any ɛ (0, 1), we have where 1 lim n n log β n(ɛ) = D(P Q) D(P Q) x X P (x) log P (x) Q(x) is the familiar Kullback-Leibler divergence encountered in information theory. Note that this result is similar to the idea of strong converse in information theory since the limit holds for all ɛ (0, 1) and not just for the case when ɛ 0. A simple proof of this lemma using typicality is sketched below. Proof. We need to show achievability of the said exponent and show that it isn t possible to improve upon it. We first show the direct part. Let the random variables X and X be distributed according to distributions P (H 0 ) and Q (H 1 ) respectively. Let the acceptance region A n be the typical set Tµ n (X). By strong law of large numbers, we have α n = P (X n A c n) ɛ if we choose µ < ɛ. Also, we have P ( X n = x n ) = exp[ n (H(P x n) + D(P x n Q))] 2
4 where P x n is the type of x n. Using this, we can calculate β n (ɛ) as β n = x A n P ( X n = x n ) = x A n exp[ n (H(P x n) + D(P x n Q))] A n max P x n:x n A n exp[ n (H(P x n) + D(P x n Q))] exp[n (H(P ) + ɛ 1 )]exp[ n (H(P ) + D(P Q))] from continuity of entropy exp[ n (D(P Q) ɛ 1 )] where ɛ 1 0 as ɛ 0 This shows the achievability of the exponent D(P Q) To show the converse part, suppose we are given an optimal acceptance region A n which minimizes β n with α n ɛ. Set B n = A n Tµ n (X). Then P (X n B n ) 1 ɛ µ which gives a bound on the cardinality of B n as B n (1 ɛ µ) exp[n (H(X) µ)]. Hence, β n = P ( X n A n ) P ( X n B n ) = x n B n P ( X n = x n ) = B n exp[ n (H(P ) + D(P Q) + δ)] where δ 0 as µ 0 (1 ɛ µ) exp[ n (D(P Q) + δ + µ)] Now, it follows from the assumed optimality of A n that the optimal β n D(P Q) + δ + µ. Since µ is arbitrary, this establishes stein s lemma. 2 Hypothesis Testing with Communication Constraints So far, we have discussed the classical version of the hypothesis testing problem. Now, we add a new dimension to this problem by assuming that the statistician does not have direct access to the data. Rather, the data is remotely collected and made available to him/her through noiseless channels of finite capacity. This is illustrated in figure 1. 3
5 n X X Encoder n f( X ) Statistician H 0 n Y Y Encoder g( Y n ) H 1 Figure 1: Hypothesis testing with two-sided compression Note that in the case of single-user hypothesis testing, the statistician requires only one bit of information (whether x n A n or not) to make an optimal decision. Hence this problem is entirely trivial. The simplest non-trivial case is when there are atleast 2 sources as illustrated in figure 1. This system was first studied by Ahlswede and Csisźar [1] in a simplified form where they considered test against independence with one-sided compression. They also derived a lower bound for the case of more general bivariate testing. This lower bound was later improved by Te Sun Han [2] whose formula subsumes all other known bounds. We shall, in this report, be concerned mainly about this lower bound. A good survey of statistical inference problems under communication constraints can be found in [3]. There are a couple of important differences between multi-terminal information theory and statistical testing with multi-terminal data compression. In the hypothesis testing problem, we are not interested in recovering the original data in any sense. We are only interested in designing the best possible test to differentiate between H 0 and H 1. The performance measure here is the rate needed to transmit the data and the best possible exponent based on that compressed data. Indeed, information theory tells us that if the rates R 1 and R 2 are not large enough, the probability of error in decoding goes to 1 exponentially. But even in such cases, it is possible to do very well in the hypothesis testing problem. In fact, even the case of (R 1, R 2 ) = (0, 0) yields a non-zero exponent in the hypothesis testing problem. (see [2]). 4
6 2.1 Bivariate Hypotheses with One-sided Compression Let s first consider a simpler case than the system in figure 1 where one of the data streams (say y n ) is assumed to be available without compression. Let the null hypothesis be H 0 = (P XY ) and the alternative hypothesis be H 1 = (P XȲ ). The main result here is to establish a good lower bound on the exponent θ n (ɛ) 1 n log β n(ɛ). Before we proceed to the main theorem, we need the following lemma. Lemma 2.1. Suppose that η > 0, δ > 0 are arbitrary fixed, and let U,X,Y be finite random variables such that U X Y. Let M = exp[n(i(u; X) + η)]. Then there exist u 1, u 2,..., u M Tµ n (U) and M disjoint subsets C 1,..., C M such that C i Tµ n (X u i ) for which M P { X n Y n C i Tµ n (Y u i ) } 1 δ (1) i=1 Proof. The lemma can be proved using standard ideas from multi-user information theory and is omitted here. It can be found in [2]. We will state the main result of this section below. Define two sets of auxiliary random variables: S(R) = {U : I(U; X) R, U X Y } L(U) = {Ũ XỸ : P (Ũ X) } = P (UX), P (ŨỸ ) = P (UY ) Also define the random variable Ū to satisfy Ū X Ȳ and P (Ū X) = P (U X). Define the function θ L (R) = sup min D(Ũ U S(R) Ũ XỸ L(U) XỸ Ū XȲ ) Then we have the following theorem Theorem 2.1. Let θ(r, ɛ) be the largest possible exponent for the case of one-sided data compression at rate R. Then, for all R 0 and 0 < ɛ < 1, θ(r, ɛ) θ L (R) 5
7 Proof. It is sufficient to show that for each U S(R) θ(r, ɛ) min D(Ũ XỸ Ū XȲ ) Ũ XỸ L(U) Let M, u 1, u 2,..., u M T n µ (U), C i T n µ (X u i ) be as specified in Lemma 2.1. The X-encoder is described as i if x n C f(x n i ) = 0 else Note that since M = exp[n(i(u; X)+η)], the rate constraint is automatically satisfied for this encoder. The statistician has access to i {1, 2,..., M} and y n with which to make a decision. The decision rule is to declare H 0 if y n Tµ n (Y u i ). Globally, this translates into an acceptance region defined by M ( A n = Ci Tµ n (Y u i ) ) (2) i=1 Note that this acceptance region A n is defined for computational purposes only and no single module in our system has all the information required to determine if indeed (x n, y n ) A n. We now compute the error probabilities α n and β n. Evaluation of α n : Clearly, by the definition of A n and from equation (1), we have α n = P (X n Y n A c n) δ Evaluation of β n : Error probability of the second kind can be evaluated as β n = = (x n,y n ) A n P (x n,y n ) A n exp ( ( Xn, Ȳ n ) = (x n, y n ) ) [ ( ( ) ( ))] n H P(x n,y n ) + D P(x n,y n ) P XȲ (3) We now convert the summation over individual sequences to summation over types. This involves estimating the number of different types P (x n,y n ) in A n and number of sequences in each 6
8 of them. Let U (n), X (n), Y (n) be the type variables of the n-length sequences u n, x n, y n respectively. Let K(X (n) Y (n) ) be the number of (x n, y n ) A n with type given by (X (n) Y (n) ). Let K i (X (n) Y (n) ) (i = 1, 2,..., M) be the number of (x n, y n ) C i T n µ (Y u i ) with type given by (X (n) Y (n) ). By the disjointed nature of C i s, we have K(X (n) Y (n) ) = M i=1 K i(x (n) Y (n) ). Using these in equation (3), we get β n = = (X (n) Y (n) ) M K i=1 (U (n) X (n) Y (n) ) ( X (n) Y (n)) [ ( ( ) ( ))] exp n H P (X (n),y (n) ) + D P (X (n),y (n) ) P XȲ K i (X (n) Y (n)) [ ( ( ) ( ))] exp n H P (X (n),y (n) ) + D P (X (n),y (n) ) P XȲ A simple upper bound on K i ( X (n) Y (n)) is given by K i (X (n) Y (n)) [ ( Tµ n (Y u i ) exp n H (X (n) U (n) Y (n)))] Using this bound and the fact that M = exp[n(i(u; X) + η)], we get where β n (U (n) X (n) Y (n) ) [ ( )] exp n d(u (n) X (n) Y (n) ) 2µ η [ ( )] (n + 1) U X Y max (U (n) X (n) Y (n) ) exp n d(u (n) X (n) Y (n) ) 2µ η (4) d(u (n) X (n) Y (n) ) H(X (n) Y (n) ) + D(X (n) Y (n) XȲ ) H(X (n) Y (n) U (n) ) H(Y U) I(U; X) (5) We now determine the allowable types (U (n) X (n) Y (n) ) in A n. If (x n, y n ) A n, then by the definition of A n, x n C i Tµ n (X u i ) and y n Tµ n (Y u i ) for some i {1, 2,..., M}. But since the u i themselves are chosen from the typical set Tµ n (U), this implies that (x n, u n ) Tµ n (X, U) and (y n, u n ) Tµ n (Y, U). Hence the type variables (U (n) X (n) ) have to be close in distribution to (U, X) and similarly (U (n) Y (n) ) must be close to (U, Y ). Using continuity of all the information theoretic quantities, we can now rewrite equation (5) as d(u (n) X (n) Y (n) ) = H( XỸ ) + D( XỸ XȲ ) H( X ŨỸ ) H(Ỹ Ũ) I(Ũ; X) + δ (6) 7
9 for some auxiliary random variables Ũ, X, Ỹ such that P Ũ X = P UX and PŨỸ = P UY. This can be further simplified as d(u (n) X (n) Y (n) ) = D(Ũ XỸ Ū XȲ ) + δ where Ū is a random variable uniquely defined through the conditions p(ū X) = p(u X) and Ū X Ȳ. Substituting this simplified form of d(u (n) X (n) Y (n) ) into equation (4), we can conclude that θ L (R, ɛ) = min D(Ũ XỸ Ū XȲ ) Ũ XỸ L(U) is an achievable lower bound to θ(r, ɛ) for every U in S(R). Thus, we have succeeded in deriving a single-letter lower bound for the power exponent in the case of hypothesis testing with one-sided data compression. The extension to the case of two-sided data compression is straight-forward and involves the introduction of another pair of auxiliary random variables V and Ṽ. The proof techniques and results are analogous to the ones discussed above. 2.2 Special Cases We will now describe several results that are easily derived from the above theorem. Corollary 1. If R H(X), then for all ɛ (0, 1) θ L (R, ɛ) = D ( XY XȲ ) Proof. If R H(X), we can choose the auxiliary random variable U S(R) to be identical with X. In that case, we have P (Ũ XỸ ) = P (UXY ) and hence θ L(R) = D(XY XȲ ). By Stein s lemma, this is also the best obtainable power exponent. Corollary 2. For any U S(R) and 0 < ɛ < 1, θ L (R, ɛ) D(X X) + D(UY UŶ ) 8
10 where Ŷ is the random variable with U X Ŷ and p(ŷ X) = p(ȳ X) This lower bound was originally derived by Ahlswede and Csisźar in [1]. Proof. In the proof, we will need to make use of the fact that for any finite random variables X 1, X 2, Y 1, Y 2 D (X 1 Y 1 X 2 Y 2 ) D (X 1 X 2 ) (7) Also note that, by the definition of Ŷ, we have P (UŶ X) = P (U X)P (Ŷ X) = P (U X)P (Ȳ X). For any U S(R) and Ũ XỸ L(U), D (Ũ XỸ Ū XȲ ) = D(X X) + u,x,y P ( Ũ XỸ )(u, x, y) log P ŨỸ X (u, y x) P ŪȲ X (u, y x) = D(X X) + u,x,y = D(X X) + u,x,y P ( Ũ XỸ )(u, x, y) log PŨ XỸ (u, x, y) P Ū X (u x)pȳ X (y x)p X(x) P ( Ũ XỸ )(u, x, y) log PŨ XỸ (u, x, y) P U Ŷ X (u, y x)p X(x) D(X X) + u,y P UY (u, y) log P UY (u, y) P U Ŷ (u, y) from equation (7) = D(X X) + D(UY UŶ ) Corollary 3. Suppose P ( XȲ ) = P (X)P (Y ). Then for any 0 < ɛ < 1, θ L (R, ɛ) max I(U; Y ) U S(R) This is the problem of testing against independence that was completely resolved in [1]. Proof. From equation (7) we have D(Ũ XỸ Ū XȲ ) D(ŨỸ ŪȲ ). Since P (ŨỸ ) = P (UY ), P (Ū) = P (U), P (Ȳ ) = P (Y ) and Ū, Ȳ are independent, this further reduces to D(ŨỸ ŪȲ ) = D(P UY P U P Y ) = I(U; Y ) Note that, in [1] the authors also provided a strong converse to the above result using a very different proof technique. (using results from [4]) 9
11 2.3 Improvement of the Power Exponent So far, we have looked at an encoding scheme that is designed to reproduce the joint types P (u n,x n ) and P (u n,y n ) exactly at the decoder with zero-error probability. One can also consider a wider class of encoders that guarantee exponentially decaying non-zero error probabilities. In this case, the source of error is two-fold and there is a tradeoff between them. The error events are Error in hypothesis testing given an acceptance region Since there is now a possibility of error in the acceptance region, the set L (U) is now enlarged to L (U) = {Ũ XỸ P (Ũ X) = P (UX), P (Ỹ ) = P (Y ), H(Ũ Ỹ ) H(U Y )} The resulting exponent is ρ 1(U) = min D(Ũ XỸ XȲ Z) Ũ XỸ Encoding error in specifying an acceptance region By allowing for this error event, the set S(R) of allowed Us can be enlarged to S (R) = {U R I(U; X Y ), U X Y } The resulting exponent is + if R I(U; X) ρ 2(U) = ρ 2 (U) otherwise where ρ 2 (U) = [R I(U; X Y ] + + min Ũ XỸ L (U) D(Ũ XỸ XȲ Z) Clearly the minimum of the 2 exponents is achievable for every U L (R) and so we have that θ L(R, ɛ) = sup min(ρ 1(U), ρ 2(U)) U L (R) is an achievable exponent. Since this scheme allows for a much wider class of encoders, this lower bound is much tighter than the previous bound given in Theorem
12 3 Conclusion Thus far, we have dealt with the multi-terminal hypothesis testing problem in the presence of onesided data compression. Two single-letter lower bounds θ L (R, ɛ) and θl (R, ɛ) were derived. However, the problem isn t completely resolved since it isn t known if the lower bounds are tight. More general hypothesis testing problems seem to be very intractable with very few single-letter characterizations available. The only fully solved model is that of testing against independence studied in [1]. The approach of divergence characterization by Ahlswede and Csisźar has formidable complexity in proving the direct part but lends itself to converses (using ideas such as the blowing-up lemma). The approach detailed here is useful for establishing achievability results but seem to be insufficient in establishing converses. Te sun Han and Amari [3] have also studied other statistical inference problems such as parameter estimation and pattern classification under similar rate constraints. References [1] R. Ahlswede and I. Csisźar, Hypothesis testing with commuication constraints, IEEE transactions on information theory, vol. IT-32, pp , July [2] T. S. Han, Hypothesis testing with multiterminal data compression, IEEE transactions on information theory, vol. IT-33, pp , November [3] T. S. Han and S.-I. Amari, Statistical inference under multiterminal data compression, IEEE transactions on information theory, vol. 44, pp , October [4] R. Ahlswede and J.Korner, Source coding with side information and a converse for degraded broadcast channels, IEEE transactions on information theory, vol. IT-21, pp , November
Hypothesis Testing with Communication Constraints
Hypothesis Testing with Communication Constraints Dinesh Krithivasan EECS 750 April 17, 2006 Dinesh Krithivasan (EECS 750) Hyp. testing with comm. constraints April 17, 2006 1 / 21 Presentation Outline
More informationInformation measures in simple coding problems
Part I Information measures in simple coding problems in this web service in this web service Source coding and hypothesis testing; information measures A(discrete)source is a sequence {X i } i= of random
More informationThe Method of Types and Its Application to Information Hiding
The Method of Types and Its Application to Information Hiding Pierre Moulin University of Illinois at Urbana-Champaign www.ifp.uiuc.edu/ moulin/talks/eusipco05-slides.pdf EUSIPCO Antalya, September 7,
More informationArimoto Channel Coding Converse and Rényi Divergence
Arimoto Channel Coding Converse and Rényi Divergence Yury Polyanskiy and Sergio Verdú Abstract Arimoto proved a non-asymptotic upper bound on the probability of successful decoding achievable by any code
More informationOn the Necessity of Binning for the Distributed Hypothesis Testing Problem
On the Necessity of Binning for the Distributed Hypothesis Testing Problem Gil Katz, Pablo Piantanida, Romain Couillet, Merouane Debbah To cite this version: Gil Katz, Pablo Piantanida, Romain Couillet,
More informationDistributed Detection With Vector Quantizer Wenwen Zhao, Student Member, IEEE, and Lifeng Lai, Member, IEEE
IEEE TRANSACTIONS ON SIGNAL AND INFORMATION PROCESSING OVER NETWORKS, VOL 2, NO 2, JUNE 206 05 Distributed Detection With Vector Quantizer Wenwen Zhao, Student Member, IEEE, and Lifeng Lai, Member, IEEE
More informationInteractive Hypothesis Testing with Communication Constraints
Fiftieth Annual Allerton Conference Allerton House, UIUC, Illinois, USA October - 5, 22 Interactive Hypothesis Testing with Communication Constraints Yu Xiang and Young-Han Kim Department of Electrical
More informationCommon Randomness Principles of Secrecy
Common Randomness Principles of Secrecy Himanshu Tyagi Department of Electrical and Computer Engineering and Institute of Systems Research 1 Correlated Data, Distributed in Space and Time Sensor Networks
More informationQuiz 2 Date: Monday, November 21, 2016
10-704 Information Processing and Learning Fall 2016 Quiz 2 Date: Monday, November 21, 2016 Name: Andrew ID: Department: Guidelines: 1. PLEASE DO NOT TURN THIS PAGE UNTIL INSTRUCTED. 2. Write your name,
More information(each row defines a probability distribution). Given n-strings x X n, y Y n we can use the absence of memory in the channel to compute
ENEE 739C: Advanced Topics in Signal Processing: Coding Theory Instructor: Alexander Barg Lecture 6 (draft; 9/6/03. Error exponents for Discrete Memoryless Channels http://www.enee.umd.edu/ abarg/enee739c/course.html
More informationLecture 2: August 31
0-704: Information Processing and Learning Fall 206 Lecturer: Aarti Singh Lecture 2: August 3 Note: These notes are based on scribed notes from Spring5 offering of this course. LaTeX template courtesy
More informationDistributed Hypothesis Testing Over Discrete Memoryless Channels
1 Distributed Hypothesis Testing Over Discrete Memoryless Channels Sreejith Sreekumar and Deniz Gündüz Imperial College London, UK Email: {s.sreekumar15, d.gunduz}@imperial.ac.uk arxiv:1802.07665v6 [cs.it]
More informationCorrelation Detection and an Operational Interpretation of the Rényi Mutual Information
Correlation Detection and an Operational Interpretation of the Rényi Mutual Information Masahito Hayashi 1, Marco Tomamichel 2 1 Graduate School of Mathematics, Nagoya University, and Centre for Quantum
More informationInformation Theory in Intelligent Decision Making
Information Theory in Intelligent Decision Making Adaptive Systems and Algorithms Research Groups School of Computer Science University of Hertfordshire, United Kingdom June 7, 2015 Information Theory
More informationLecture 22: Final Review
Lecture 22: Final Review Nuts and bolts Fundamental questions and limits Tools Practical algorithms Future topics Dr Yao Xie, ECE587, Information Theory, Duke University Basics Dr Yao Xie, ECE587, Information
More informationSHARED INFORMATION. Prakash Narayan with. Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe
SHARED INFORMATION Prakash Narayan with Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe 2/40 Acknowledgement Praneeth Boda Himanshu Tyagi Shun Watanabe 3/40 Outline Two-terminal model: Mutual
More informationNetwork coding for multicast relation to compression and generalization of Slepian-Wolf
Network coding for multicast relation to compression and generalization of Slepian-Wolf 1 Overview Review of Slepian-Wolf Distributed network compression Error exponents Source-channel separation issues
More informationDistributed Hypothesis Testing Over Discrete Memoryless Channels
1 Distributed Hypothesis Testing Over Discrete Memoryless Channels Sreejith Sreekumar and Deniz Gündüz Imperial College London, UK Email: {s.sreekumar15, d.gunduz}@imperial.ac.uk Abstract A distributed
More informationLattices for Distributed Source Coding: Jointly Gaussian Sources and Reconstruction of a Linear Function
Lattices for Distributed Source Coding: Jointly Gaussian Sources and Reconstruction of a Linear Function Dinesh Krithivasan and S. Sandeep Pradhan Department of Electrical Engineering and Computer Science,
More informationLECTURE 10. Last time: Lecture outline
LECTURE 10 Joint AEP Coding Theorem Last time: Error Exponents Lecture outline Strong Coding Theorem Reading: Gallager, Chapter 5. Review Joint AEP A ( ɛ n) (X) A ( ɛ n) (Y ) vs. A ( ɛ n) (X, Y ) 2 nh(x)
More informationIntroduction to Information Theory. Uncertainty. Entropy. Surprisal. Joint entropy. Conditional entropy. Mutual information.
L65 Dept. of Linguistics, Indiana University Fall 205 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission rate
More informationDept. of Linguistics, Indiana University Fall 2015
L645 Dept. of Linguistics, Indiana University Fall 2015 1 / 28 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission
More informationSuperposition Encoding and Partial Decoding Is Optimal for a Class of Z-interference Channels
Superposition Encoding and Partial Decoding Is Optimal for a Class of Z-interference Channels Nan Liu and Andrea Goldsmith Department of Electrical Engineering Stanford University, Stanford CA 94305 Email:
More informationLecture 21: Minimax Theory
Lecture : Minimax Theory Akshay Krishnamurthy akshay@cs.umass.edu November 8, 07 Recap In the first part of the course, we spent the majority of our time studying risk minimization. We found many ways
More informationEE/Stat 376B Handout #5 Network Information Theory October, 14, Homework Set #2 Solutions
EE/Stat 376B Handout #5 Network Information Theory October, 14, 014 1. Problem.4 parts (b) and (c). Homework Set # Solutions (b) Consider h(x + Y ) h(x + Y Y ) = h(x Y ) = h(x). (c) Let ay = Y 1 + Y, where
More informationCapacity of a channel Shannon s second theorem. Information Theory 1/33
Capacity of a channel Shannon s second theorem Information Theory 1/33 Outline 1. Memoryless channels, examples ; 2. Capacity ; 3. Symmetric channels ; 4. Channel Coding ; 5. Shannon s second theorem,
More informationSHARED INFORMATION. Prakash Narayan with. Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe
SHARED INFORMATION Prakash Narayan with Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi, Shun Watanabe 2/41 Outline Two-terminal model: Mutual information Operational meaning in: Channel coding: channel
More information4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information
4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information Ramji Venkataramanan Signal Processing and Communications Lab Department of Engineering ramji.v@eng.cam.ac.uk
More informationThe Capacity Region for Multi-source Multi-sink Network Coding
The Capacity Region for Multi-source Multi-sink Network Coding Xijin Yan Dept. of Electrical Eng. - Systems University of Southern California Los Angeles, CA, U.S.A. xyan@usc.edu Raymond W. Yeung Dept.
More informationDispersion of the Gilbert-Elliott Channel
Dispersion of the Gilbert-Elliott Channel Yury Polyanskiy Email: ypolyans@princeton.edu H. Vincent Poor Email: poor@princeton.edu Sergio Verdú Email: verdu@princeton.edu Abstract Channel dispersion plays
More informationSolutions to Homework Set #2 Broadcast channel, degraded message set, Csiszar Sum Equality
1st Semester 2010/11 Solutions to Homework Set #2 Broadcast channel, degraded message set, Csiszar Sum Equality 1. Convexity of capacity region of broadcast channel. Let C R 2 be the capacity region of
More informationECE Information theory Final (Fall 2008)
ECE 776 - Information theory Final (Fall 2008) Q.1. (1 point) Consider the following bursty transmission scheme for a Gaussian channel with noise power N and average power constraint P (i.e., 1/n X n i=1
More informationFrans M.J. Willems. Authentication Based on Secret-Key Generation. Frans M.J. Willems. (joint work w. Tanya Ignatenko)
Eindhoven University of Technology IEEE EURASIP Spain Seminar on Signal Processing, Communication and Information Theory, Universidad Carlos III de Madrid, December 11, 2014 : Secret-Based Authentication
More informationAN INTRODUCTION TO SECRECY CAPACITY. 1. Overview
AN INTRODUCTION TO SECRECY CAPACITY BRIAN DUNN. Overview This paper introduces the reader to several information theoretic aspects of covert communications. In particular, it discusses fundamental limits
More informationThe Gallager Converse
The Gallager Converse Abbas El Gamal Director, Information Systems Laboratory Department of Electrical Engineering Stanford University Gallager s 75th Birthday 1 Information Theoretic Limits Establishing
More informationLecture 11: Continuous-valued signals and differential entropy
Lecture 11: Continuous-valued signals and differential entropy Biology 429 Carl Bergstrom September 20, 2008 Sources: Parts of today s lecture follow Chapter 8 from Cover and Thomas (2007). Some components
More informationINFORMATION THEORY AND STATISTICS
CHAPTER INFORMATION THEORY AND STATISTICS We now explore the relationship between information theory and statistics. We begin by describing the method of types, which is a powerful technique in large deviation
More informationChapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University
Chapter 4 Data Transmission and Channel Capacity Po-Ning Chen, Professor Department of Communications Engineering National Chiao Tung University Hsin Chu, Taiwan 30050, R.O.C. Principle of Data Transmission
More informationLecture 10: Broadcast Channel and Superposition Coding
Lecture 10: Broadcast Channel and Superposition Coding Scribed by: Zhe Yao 1 Broadcast channel M 0M 1M P{y 1 y x} M M 01 1 M M 0 The capacity of the broadcast channel depends only on the marginal conditional
More informationEE5139R: Problem Set 4 Assigned: 31/08/16, Due: 07/09/16
EE539R: Problem Set 4 Assigned: 3/08/6, Due: 07/09/6. Cover and Thomas: Problem 3.5 Sets defined by probabilities: Define the set C n (t = {x n : P X n(x n 2 nt } (a We have = P X n(x n P X n(x n 2 nt
More informationLecture 6: Gaussian Channels. Copyright G. Caire (Sample Lectures) 157
Lecture 6: Gaussian Channels Copyright G. Caire (Sample Lectures) 157 Differential entropy (1) Definition 18. The (joint) differential entropy of a continuous random vector X n p X n(x) over R is: Z h(x
More informationArimoto-Rényi Conditional Entropy. and Bayesian M-ary Hypothesis Testing. Abstract
Arimoto-Rényi Conditional Entropy and Bayesian M-ary Hypothesis Testing Igal Sason Sergio Verdú Abstract This paper gives upper and lower bounds on the minimum error probability of Bayesian M-ary hypothesis
More informationECE 4400:693 - Information Theory
ECE 4400:693 - Information Theory Dr. Nghi Tran Lecture 8: Differential Entropy Dr. Nghi Tran (ECE-University of Akron) ECE 4400:693 Lecture 1 / 43 Outline 1 Review: Entropy of discrete RVs 2 Differential
More informationVariable Length Codes for Degraded Broadcast Channels
Variable Length Codes for Degraded Broadcast Channels Stéphane Musy School of Computer and Communication Sciences, EPFL CH-1015 Lausanne, Switzerland Email: stephane.musy@ep.ch Abstract This paper investigates
More informationOn Large Deviation Analysis of Sampling from Typical Sets
Communications and Signal Processing Laboratory (CSPL) Technical Report No. 374, University of Michigan at Ann Arbor, July 25, 2006. On Large Deviation Analysis of Sampling from Typical Sets Dinesh Krithivasan
More informationLarge Deviations Performance of Knuth-Yao algorithm for Random Number Generation
Large Deviations Performance of Knuth-Yao algorithm for Random Number Generation Akisato KIMURA akisato@ss.titech.ac.jp Tomohiko UYEMATSU uematsu@ss.titech.ac.jp April 2, 999 No. AK-TR-999-02 Abstract
More informationDistributed Functional Compression through Graph Coloring
Distributed Functional Compression through Graph Coloring Vishal Doshi, Devavrat Shah, Muriel Médard, and Sidharth Jaggi Laboratory for Information and Decision Systems Massachusetts Institute of Technology
More informationStrong Converse and Stein s Lemma in the Quantum Hypothesis Testing
Strong Converse and Stein s Lemma in the Quantum Hypothesis Testing arxiv:uant-ph/9906090 v 24 Jun 999 Tomohiro Ogawa and Hiroshi Nagaoka Abstract The hypothesis testing problem of two uantum states is
More informationQuantum Achievability Proof via Collision Relative Entropy
Quantum Achievability Proof via Collision Relative Entropy Salman Beigi Institute for Research in Fundamental Sciences (IPM) Tehran, Iran Setemper 8, 2014 Based on a joint work with Amin Gohari arxiv:1312.3822
More informationIntermittent Communication
Intermittent Communication Mostafa Khoshnevisan, Student Member, IEEE, and J. Nicholas Laneman, Senior Member, IEEE arxiv:32.42v2 [cs.it] 7 Mar 207 Abstract We formulate a model for intermittent communication
More informationSolutions to Homework Set #1 Sanov s Theorem, Rate distortion
st Semester 00/ Solutions to Homework Set # Sanov s Theorem, Rate distortion. Sanov s theorem: Prove the simple version of Sanov s theorem for the binary random variables, i.e., let X,X,...,X n be a sequence
More informationSecret Key Agreement: General Capacity and Second-Order Asymptotics. Masahito Hayashi Himanshu Tyagi Shun Watanabe
Secret Key Agreement: General Capacity and Second-Order Asymptotics Masahito Hayashi Himanshu Tyagi Shun Watanabe Two party secret key agreement Maurer 93, Ahlswede-Csiszár 93 X F Y K x K y ArandomvariableK
More informationLecture 8: Information Theory and Statistics
Lecture 8: Information Theory and Statistics Part II: Hypothesis Testing and I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw December 23, 2015 1 / 50 I-Hsiang
More informationOn the Duality between Multiple-Access Codes and Computation Codes
On the Duality between Multiple-Access Codes and Computation Codes Jingge Zhu University of California, Berkeley jingge.zhu@berkeley.edu Sung Hoon Lim KIOST shlim@kiost.ac.kr Michael Gastpar EPFL michael.gastpar@epfl.ch
More informationLecture 22: Error exponents in hypothesis testing, GLRT
10-704: Information Processing and Learning Spring 2012 Lecture 22: Error exponents in hypothesis testing, GLRT Lecturer: Aarti Singh Scribe: Aarti Singh Disclaimer: These notes have not been subjected
More informationInformation Theory and Hypothesis Testing
Summer School on Game Theory and Telecommunications Campione, 7-12 September, 2014 Information Theory and Hypothesis Testing Mauro Barni University of Siena September 8 Review of some basic results linking
More informationA proof of the existence of good nested lattices
A proof of the existence of good nested lattices Dinesh Krithivasan and S. Sandeep Pradhan July 24, 2007 1 Introduction We show the existence of a sequence of nested lattices (Λ (n) 1, Λ(n) ) with Λ (n)
More informationLarge Deviations Performance of Interval Algorithm for Random Number Generation
Large Deviations Performance of Interval Algorithm for Random Number Generation Akisato KIMURA akisato@ss.titech.ac.jp Tomohiko UYEMATSU uematsu@ss.titech.ac.jp February 22, 999 No. AK-TR-999-0 Abstract
More informationOn Function Computation with Privacy and Secrecy Constraints
1 On Function Computation with Privacy and Secrecy Constraints Wenwen Tu and Lifeng Lai Abstract In this paper, the problem of function computation with privacy and secrecy constraints is considered. The
More informationLECTURE 15. Last time: Feedback channel: setting up the problem. Lecture outline. Joint source and channel coding theorem
LECTURE 15 Last time: Feedback channel: setting up the problem Perfect feedback Feedback capacity Data compression Lecture outline Joint source and channel coding theorem Converse Robustness Brain teaser
More informationA Graph-based Framework for Transmission of Correlated Sources over Multiple Access Channels
A Graph-based Framework for Transmission of Correlated Sources over Multiple Access Channels S. Sandeep Pradhan a, Suhan Choi a and Kannan Ramchandran b, a {pradhanv,suhanc}@eecs.umich.edu, EECS Dept.,
More informationHomework Assignment #2 for Prob-Stats, Fall 2018 Due date: Monday, October 22, 2018
Homework Assignment #2 for Prob-Stats, Fall 2018 Due date: Monday, October 22, 2018 Topics: consistent estimators; sub-σ-fields and partial observations; Doob s theorem about sub-σ-field measurability;
More informationLecture 5 Channel Coding over Continuous Channels
Lecture 5 Channel Coding over Continuous Channels I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw November 14, 2014 1 / 34 I-Hsiang Wang NIT Lecture 5 From
More informationINFORMATION PROCESSING ABILITY OF BINARY DETECTORS AND BLOCK DECODERS. Michael A. Lexa and Don H. Johnson
INFORMATION PROCESSING ABILITY OF BINARY DETECTORS AND BLOCK DECODERS Michael A. Lexa and Don H. Johnson Rice University Department of Electrical and Computer Engineering Houston, TX 775-892 amlexa@rice.edu,
More informationCapacity Region of Reversely Degraded Gaussian MIMO Broadcast Channel
Capacity Region of Reversely Degraded Gaussian MIMO Broadcast Channel Jun Chen Dept. of Electrical and Computer Engr. McMaster University Hamilton, Ontario, Canada Chao Tian AT&T Labs-Research 80 Park
More informationAnalyzing Large Communication Networks
Analyzing Large Communication Networks Shirin Jalali joint work with Michelle Effros and Tracey Ho Dec. 2015 1 The gap Fundamental questions: i. What is the best achievable performance? ii. How to communicate
More informationCapacity of the Discrete Memoryless Energy Harvesting Channel with Side Information
204 IEEE International Symposium on Information Theory Capacity of the Discrete Memoryless Energy Harvesting Channel with Side Information Omur Ozel, Kaya Tutuncuoglu 2, Sennur Ulukus, and Aylin Yener
More informationSecret Key Generation and Secure Computing
Secret Key Generation and Secure Computing Himanshu Tyagi Department of Electrical and Computer Engineering and Institute of System Research University of Maryland, College Park, USA. Joint work with Prakash
More informationOn Common Information and the Encoding of Sources that are Not Successively Refinable
On Common Information and the Encoding of Sources that are Not Successively Refinable Kumar Viswanatha, Emrah Akyol, Tejaswi Nanjundaswamy and Kenneth Rose ECE Department, University of California - Santa
More informationInformation Theory CHAPTER. 5.1 Introduction. 5.2 Entropy
Haykin_ch05_pp3.fm Page 207 Monday, November 26, 202 2:44 PM CHAPTER 5 Information Theory 5. Introduction As mentioned in Chapter and reiterated along the way, the purpose of a communication system is
More informationEE376A: Homeworks #4 Solutions Due on Thursday, February 22, 2018 Please submit on Gradescope. Start every question on a new page.
EE376A: Homeworks #4 Solutions Due on Thursday, February 22, 28 Please submit on Gradescope. Start every question on a new page.. Maximum Differential Entropy (a) Show that among all distributions supported
More informationInformation Masking and Amplification: The Source Coding Setting
202 IEEE International Symposium on Information Theory Proceedings Information Masking and Amplification: The Source Coding Setting Thomas A. Courtade Department of Electrical Engineering University of
More informationSOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003
SOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003 SLEPIAN-WOLF RESULT { X i} RATE R x ENCODER 1 DECODER X i V i {, } { V i} ENCODER 0 RATE R v Problem: Determine R, the
More informationThe sequential decoding metric for detection in sensor networks
The sequential decoding metric for detection in sensor networks B. Narayanaswamy, Yaron Rachlin, Rohit Negi and Pradeep Khosla Department of ECE Carnegie Mellon University Pittsburgh, PA, 523 Email: {bnarayan,rachlin,negi,pkk}@ece.cmu.edu
More informationEntropies & Information Theory
Entropies & Information Theory LECTURE I Nilanjana Datta University of Cambridge,U.K. See lecture notes on: http://www.qi.damtp.cam.ac.uk/node/223 Quantum Information Theory Born out of Classical Information
More informationLecture 6 I. CHANNEL CODING. X n (m) P Y X
6- Introduction to Information Theory Lecture 6 Lecturer: Haim Permuter Scribe: Yoav Eisenberg and Yakov Miron I. CHANNEL CODING We consider the following channel coding problem: m = {,2,..,2 nr} Encoder
More informationDistributed Structures, Sequential Optimization, and Quantization for Detection
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL., NO., JANUARY Distributed Structures, Sequential Optimization, and Quantization for Detection Michael A. Lexa, Student Member, IEEE and Don H. Johnson, Fellow,
More informationDefinition 3.1 A statistical hypothesis is a statement about the unknown values of the parameters of the population distribution.
Hypothesis Testing Definition 3.1 A statistical hypothesis is a statement about the unknown values of the parameters of the population distribution. Suppose the family of population distributions is indexed
More informationLECTURE 13. Last time: Lecture outline
LECTURE 13 Last time: Strong coding theorem Revisiting channel and codes Bound on probability of error Error exponent Lecture outline Fano s Lemma revisited Fano s inequality for codewords Converse to
More informationECE Information theory Final
ECE 776 - Information theory Final Q1 (1 point) We would like to compress a Gaussian source with zero mean and variance 1 We consider two strategies In the first, we quantize with a step size so that the
More informationInteractive Decoding of a Broadcast Message
In Proc. Allerton Conf. Commun., Contr., Computing, (Illinois), Oct. 2003 Interactive Decoding of a Broadcast Message Stark C. Draper Brendan J. Frey Frank R. Kschischang University of Toronto Toronto,
More informationNational University of Singapore Department of Electrical & Computer Engineering. Examination for
National University of Singapore Department of Electrical & Computer Engineering Examination for EE5139R Information Theory for Communication Systems (Semester I, 2014/15) November/December 2014 Time Allowed:
More informationMGMT 69000: Topics in High-dimensional Data Analysis Falll 2016
MGMT 69000: Topics in High-dimensional Data Analysis Falll 2016 Lecture 14: Information Theoretic Methods Lecturer: Jiaming Xu Scribe: Hilda Ibriga, Adarsh Barik, December 02, 2016 Outline f-divergence
More informationUniversal Anytime Codes: An approach to uncertain channels in control
Universal Anytime Codes: An approach to uncertain channels in control paper by Stark Draper and Anant Sahai presented by Sekhar Tatikonda Wireless Foundations Department of Electrical Engineering and Computer
More informationRemote Source Coding with Two-Sided Information
Remote Source Coding with Two-Sided Information Basak Guler Ebrahim MolavianJazi Aylin Yener Wireless Communications and Networking Laboratory Department of Electrical Engineering The Pennsylvania State
More informationLecture 7 Introduction to Statistical Decision Theory
Lecture 7 Introduction to Statistical Decision Theory I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw December 20, 2016 1 / 55 I-Hsiang Wang IT Lecture 7
More informationUnified Scaling of Polar Codes: Error Exponent, Scaling Exponent, Moderate Deviations, and Error Floors
Unified Scaling of Polar Codes: Error Exponent, Scaling Exponent, Moderate Deviations, and Error Floors Marco Mondelli, S. Hamed Hassani, and Rüdiger Urbanke Abstract Consider transmission of a polar code
More informationIN [1], Forney derived lower bounds on the random coding
Exact Random Coding Exponents for Erasure Decoding Anelia Somekh-Baruch and Neri Merhav Abstract Random coding of channel decoding with an erasure option is studied By analyzing the large deviations behavior
More informationGeneralized Neyman Pearson optimality of empirical likelihood for testing parameter hypotheses
Ann Inst Stat Math (2009) 61:773 787 DOI 10.1007/s10463-008-0172-6 Generalized Neyman Pearson optimality of empirical likelihood for testing parameter hypotheses Taisuke Otsu Received: 1 June 2007 / Revised:
More informationAn Achievable Error Exponent for the Mismatched Multiple-Access Channel
An Achievable Error Exponent for the Mismatched Multiple-Access Channel Jonathan Scarlett University of Cambridge jms265@camacuk Albert Guillén i Fàbregas ICREA & Universitat Pompeu Fabra University of
More informationarxiv: v1 [cs.it] 19 Aug 2008
Distributed Source Coding using Abelian Group Codes arxiv:0808.2659v1 [cs.it] 19 Aug 2008 Dinesh Krithivasan and S. Sandeep Pradhan, Department of Electrical Engineering and Computer Science, University
More informationLecture 8: Information Theory and Statistics
Lecture 8: Information Theory and Statistics Part II: Hypothesis Testing and Estimation I-Hsiang Wang Department of Electrical Engineering National Taiwan University ihwang@ntu.edu.tw December 22, 2015
More informationPolar Codes are Optimal for Write-Efficient Memories
Polar Codes are Optimal for Write-Efficient Memories Qing Li Department of Computer Science and Engineering Texas A & M University College Station, TX 7784 qingli@cse.tamu.edu Anxiao (Andrew) Jiang Department
More informationAdvanced Topics in Information Theory
Advanced Topics in Information Theory Lecture Notes Stefan M. Moser c Copyright Stefan M. Moser Signal and Information Processing Lab ETH Zürich Zurich, Switzerland Institute of Communications Engineering
More informationUpper Bounds on the Capacity of Binary Intermittent Communication
Upper Bounds on the Capacity of Binary Intermittent Communication Mostafa Khoshnevisan and J. Nicholas Laneman Department of Electrical Engineering University of Notre Dame Notre Dame, Indiana 46556 Email:{mhoshne,
More informationKeyless authentication in the presence of a simultaneously transmitting adversary
Keyless authentication in the presence of a simultaneously transmitting adversary Eric Graves Army Research Lab Adelphi MD 20783 U.S.A. ericsgra@ufl.edu Paul Yu Army Research Lab Adelphi MD 20783 U.S.A.
More informationUniversal Incremental Slepian-Wolf Coding
Proceedings of the 43rd annual Allerton Conference, Monticello, IL, September 2004 Universal Incremental Slepian-Wolf Coding Stark C. Draper University of California, Berkeley Berkeley, CA, 94720 USA sdraper@eecs.berkeley.edu
More informationOn the Rate-Limited Gelfand-Pinsker Problem
On the Rate-Limited Gelfand-Pinsker Problem Ravi Tandon Sennur Ulukus Department of Electrical and Computer Engineering University of Maryland, College Park, MD 74 ravit@umd.edu ulukus@umd.edu Abstract
More informationHands-On Learning Theory Fall 2016, Lecture 3
Hands-On Learning Theory Fall 016, Lecture 3 Jean Honorio jhonorio@purdue.edu 1 Information Theory First, we provide some information theory background. Definition 3.1 (Entropy). The entropy of a discrete
More informationMultiterminal Source Coding with an Entropy-Based Distortion Measure
Multiterminal Source Coding with an Entropy-Based Distortion Measure Thomas Courtade and Rick Wesel Department of Electrical Engineering University of California, Los Angeles 4 August, 2011 IEEE International
More information